Unverified50% confidenceFactExact time
Terminal-Bench 2.1 evaluates an AI's ability to complete tasks in a command-line terminal environment
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTerminal-Bench-Science is the science branch of the Terminal-Bench family.60% similarUnverifiedIn Terminal-Bench Evolve, each task keeps its goal unchanged but continuously modifies the execution conditions, deriving five environment states from environment 0 through 4.58% similarUnverified2Tickets强制AI按「用户功能」（垂直切片）而非技术架构拆分任务，使每完成一个小功能即可立即测试57% similarUnverifiedThe general-purpose Run Command tool has an overly broad description, which may cause AI to go off track and modify things it should not touch, such as the skybox.57% similarUnverifiedAIVenture's second level demonstrates an agent's autonomous thinking loop, including evaluating game state, executing tool calls to locate hidden switches, and interacting with the Phaser.js engine to solve puzzles.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931492API
curl https://kongchang.com/api/v1/knowledge/claims/931492MCP
get_claim(id=931492)